PrimeGrid Challenge Series 2026

[Skillz]

Can't get my 9950X to accept any changes no matter what I have tried...still running one task on 16 cores. The 9900X's are acting the same as well.

Did you downgrade those BOINC installs also?

Are you using process lasso to pin the tasks to physical cores?

 

[Rollo]

Did you downgrade those BOINC installs also?

Are you using process lasso to pin the tasks to physical cores?

No, I did not reinstall BOINC on any other machines. I also have not installed Lasso on any other machines either...I'll try it on the 9950X.

 

[Skillz]

All hosts need 8.0.4 or below. Otherwise they will all fail to adhere to the correct thread settings.

Process lasso is an absolute must. Without it and you're tasks will be extremely slow. (Or you can use the power shell script)

Either way, you absolutely need to be pinning these tasks to physical cores.

 

[Rollo]

It would be more than aggravating to do that now on all my machines. I'll just wait until the end of the challenge to address that. As old as BOINC is why is it still necessary to jump through all these hoops? We should be able to assign different projects to different machines at the same time and not have to rely on a catch all preference for all of them.

 

[Skillz]

It would be more than aggravating to do that now on all my machines. I'll just wait until the end of the challenge to address that. As old as BOINC is why is it still necessary to jump through all these hoops? We should be able to assign different projects to different machines at the same time and not have to rely on a catch all preference for all of them.

Unlike F@H, BOINC consists of 30+ "F@H" projects that all have their own optimizations. Primegrid just happens to be one that requires external software to help the tasks run more optimal.

Once Process Lasso is installed (this is windows only by the way) then you just need to adjust the settings depending on which PG sub project you want to run. Basically just changing the number of cores to pin to the tasks as not all of them need 8 cores.

It's actually an OS problem and not a BOINC problem. Windows isn't smart enough to know it needs to keep the tasks all pinned on specific CCD/CCX on AMD or on specific CPU sockets for multiple socket Intel rigs.

Linux is a little better at it, but still not as aggressive as it needs to be for the PG apps.

I think people have tested pinning f@h CPU tasks as well, but since I don't follow that project very much I don't know that it gives any benefit.

 

[Rollo]

People only run F@H on a cpu as a last resort...not worth the energy expended.

 

[StefanR5R]

I will write more later.

Hadn't had time for this yesterday, may not have time today either... Also, I wish I had a dual-CCX Ryzen myself, then I would have posted precise numbers in terms of Points Per Day from measurements of a stock client <--> halfways optimized client <--> fully optimized client, so that everyone could judge for themselves whether or not it is worth to them personally to try and replicate parts or all of the optimizations.

BTW @Rollo, until 2017, PrimeGrid's applications were only singlethreaded. So the optimization aspects which I enumerated earlier (On a given host: 1. How many tasks should run concurrently? 2. Should or shouldn't Hyperthreading/SMT be used? 3. Does the OS need help with CPU scheduling?) were a lot easier to address because of lack of opportunities... and performance was much poorer than today.

Most other BOINC projects have only single-threaded applications too, hence less optimization opportunities = less optimization headaches. :-) And what's more, PrimeGrid is the one project at which contests are held regularly, hence much more motivation to optimize the heck out of it compared with other BOINC projects.

 

[mmonnin03]

Just use an app_config.xml (save to project PG folder) to force the # of threads per task vs waiting on new tasks from PG or downgrading BOINC Versions

<app_config>
<app_version>
<app_name>llrSOB</app_name>
<plan_class>mt</plan_class>
<cmdline>-t 8</cmdline>
<avg_ncpus>8</avg_ncpus>
</app_version>
</app_config>

 

[Ken g6]

Day 5.5 stats:

Rank___Credits____Username
5______21010370___cellarnoise2
6______17950749___crashtech
9______14824663___Icecold
14_____12281134___w a h
17_____10837837___ChelseaOilman
37_____4100808____Orange Kid
42_____3721553____mmonnin
46_____3495388____Rollo
58_____2471450____johnnevermind
86_____1444438____waffleironhead
129____743582___10esseeTony
140____595081_____Ken_g6
178____305718_____[TA]Skillz

Rank__Credits____Team
1_____92038224___TeAm AnandTech
2_____85472021___Czech National Team
3_____62074561___BOINC@AUSTRALIA
4_____54853065___SETI.Germany

Over halfway!

 

[cellarnoise]

It would be more than aggravating to do that now on all my machines. I'll just wait until the end of the challenge to address that. As old as BOINC is why is it still necessary to jump through all these hoops? We should be able to assign different projects to different machines at the same time and not have to rely on a catch all preference for all of them.

Optimizing multi-threaded able projects / tasks to hardware can really speed up completion times! As has been said, more and more Boinc Projects are going multi-threaded and the OS are not great at optimizing based upon the ever newer CPU architectures, which look to be getting even more complex over time.

Even optimizing for F@H cpu tasks (which can help the project, because I think they are still different work being produced than the GPU tasks - though still much less in Points compared to GPU tasks...) it is important to optimize for the CPU architecture being used as they are multi-threaded also.

 

[Rollo]

TeAm AnandTech continues to rise despite my obvious incompetence lol... 😉 I have managed to squeak in a little Folding at the same time.

 

[Skillz]

TeAm AnandTech continues to rise despite my obvious incompetence lol... 😉 I have managed to squeak in a little Folding at the same time.

Running BOINC regularly is easy.

Running BOINC competitively is a little bit more involved. 😉

 

[Ken g6]

Day 6.5 stats:

Rank___Credits____Username
5______31488850___cellarnoise2
6______23896896___Icecold
7______21469743___crashtech
13_____15552116___w a h
18_____10837837___ChelseaOilman
38_____4840346____Orange Kid
40_____4669307____Rollo
44_____4300826____mmonnin
46_____4157919___10esseeTony
60_____2902456____johnnevermind
72_____2341547____[TA]Skillz
91_____1739503____waffleironhead
147____736742_____Ken_g6

Rank__Credits____Team
1_____127189540___TeAm AnandTech
2_____104257241___Czech National Team
3_____80006147___BOINC@AUSTRALIA
4_____67903528___SETI.Germany

 

[Ken g6]

Day 7.5 stats:

Rank___Credits____Username
5______33845071___cellarnoise2
6______32082561___Icecold
8______24713364___crashtech
10_____20269533___w a h
19_____10858503___ChelseaOilman
39_____5405834____Rollo
41_____5290634____Orange Kid
43_____5026329___10esseeTony
50_____4300826____mmonnin
52_____3996808____[TA]Skillz
63_____3206011____johnnevermind
92_____2035850____waffleironhead
160____736742_____Ken_g6

Rank__Credits____Team
1_____150023517___TeAm AnandTech
2_____121512002___Czech National Team
3_____94848118___BOINC@AUSTRALIA
4_____79933497___SETI.Germany

 

[Rollo]

Romania...pretty impressive considering there are only two individuals.

 

[Rollo]

Lasso was only good for 3 days with the trial version. Any of the upcoming challenges warrant purchasing a bulk subscription for multiple machines?

 

[TennesseeTony]

If you buy it once, the key is good for several machines, but eventually they will catch on if you go too crazy with it. I think I got 'caught' on machine number 11. They have another program now I see, that seems to be free....which does what we want, let me look up the name again...eh, nevermind, it is called core parking, C6 power saving state.

Even though the trial has expired, the rules you implemented will stay in effect, you just get a VERY ANNOYING nag screen upon reboot.

The best idea might be to download and install the trial, 2.) Run every PrimeGrid project and make a rule for each, before the trial ends. 🙂 Pain in the butt.

 

[TennesseeTony]

That @Rollo fellow is hard to catch, but, I think I can, I think I can, I think I can.... 🙂

@Ken g6 hurry up with the stats...getting realllll sleepy... 😛

 

[Rollo]

If you buy it once, the key is good for several machines, but eventually they will catch on if you go too crazy with it. I think I got 'caught' on machine number 11. They have another program now I see, that seems to be free....which does what we want, let me look up the name again...eh, nevermind, it is called core parking, C6 power saving state.

Even though the trial has expired, the rules you implemented will stay in effect, you just get a VERY ANNOYING nag screen upon reboot.

The best idea might be to download and install the trial, 2.) Run every PrimeGrid project and make a rule for each, before the trial ends. 🙂 Pain in the butt.

They have a deal right now for $22 annually or $36 lifetime that covers 5 machines. Lifetime seems to be a no brainer. 5 machines would be a bit short for me, but probably enough for my purposes.

Are these upcoming challenges multi-threaded as well?

 

[Ken g6]

Day 8.5 stats:

Rank___Credits____Username
5______38203975___Icecold
6______37788953___cellarnoise2
8______28372323___crashtech
13_____21007211___w a h
17_____15693849___ChelseaOilman
35_____6710805____Rollo
38_____6626783___10esseeTony
42_____6196682____Orange Kid
54_____4300826____mmonnin
55_____4139368____[TA]Skillz
61_____3786434____johnnevermind
93_____2332487____waffleironhead
160____879003_____Ken_g6

Rank__Credits____Team
1_____174294152___TeAm AnandTech
2_____141237487___Czech National Team
3_____110484864___BOINC@AUSTRALIA
4_____94518120___SETI.Germany

Are these upcoming challenges multi-threaded as well?

I think almost all of them are, bar PPSE. The GFN ones also use GPUs, but can be run multi-threaded on CPUs as well.

 

[cellarnoise]

They have a deal right now for $22 annually or $36 lifetime that covers 5 machines. Lifetime seems to be a no brainer. 5 machines would be a bit short for me, but probably enough for my purposes.

Are these upcoming challenges multi-threaded as well?

Most of the PG challenges are mostly CPU based, but a few are just GPU and a few are both. I bought a 5 puter license of Process Lasso a few years ago and have ran it on a few more than 5, like TTony said and it has not complained. They seem to maintain the software well, and also take feedback about desired new features into future updates within a decent time frame.

You can look up what future projects and their compute type from the following link. https://www.primegrid.com/challenge/challenge.php

Also, here is a link to the overall 2026 PrimeGrid challenge's overall results:

PrimeGrid

And here are two links to what I think are fun like every 15 minute stat updates sites for past and future challenges going back a few years.

Overall Challenge Stats

Seti Germany's excellent PG Stats site for these challenges:

PG-Challenge-App (SETI.Germany)

www.seti-germany.de

The BOINC Pent is coming up also! 🙂

BOINC Pentathlon | seti-germany.de

www.seti-germany.de

 

[Skillz]

They have a deal right now for $22 annually or $36 lifetime that covers 5 machines. Lifetime seems to be a no brainer. 5 machines would be a bit short for me, but probably enough for my purposes.

Are these upcoming challenges multi-threaded as well?

Save your money and just use power shell scripts that have been posted.

 

[cellarnoise]

Save your money and just use power shell scripts that have been posted.

Those work great also, but trying to figure out what the different multi-threaded projects need and how to configure it all with different number of cores offered can be difficult between Intel and AMD, as I have figured out over the past 2 years... And then there is the thing about also having to figure out about Linux systems also, but I don't think Process Lasso is going there yet either.

 

[StefanR5R]

PS ...

Here is one of several sources for simple scripts which pin Genefer tasks to those CPUs which belong to the same CCX:
xii5ku_AT in SETI.Germany's forum on December 17, 2023
Save the Windows script with a name like "gfn_affinity.ps1" (*.ps1 is for PowerShell scripts), edit the $cores=... line according to your CPU and according to whether or not the SMT threads shall be used, and then let it execute in parallel with BOINC.

... on Linux, I use a bigger script with some more features. One simple but IMO effective feature among these is a logging function: You get to see what the script is doing. IMO this is good for the peace of mind of the user.

And on Windows, I don't use BOINC at all anymore. Therefore I never had the motivation to extend pschoefer's powershell script similarly.

Other BOINC/PrimeGrid community members have come up with other tools, notably for Windows and of course free for everybody to use, notably AffinityWatcher by Pavel Atnashev. (Sorry, I don't have a link to proper documentation of AffinityWatcher right now.)

For 99.9% of our use cases of running multithreaded applications of BOINC projects on CPUs with split last-level caches, I could see ways to make the bash script more straightforward to use. For example, it could figure out itself how the logical CPUs are organized into cache domains. I don't know how such a feature could be scripted on Windows though, except with sysinternals coreinfo as helper program.

 

[Skillz]

Those work great also, but trying to figure out what the different multi-threaded projects need and how to configure it all with different number of cores offered can be difficult between Intel and AMD, as I have figured out over the past 2 years... And then there is the thing about also having to figure out about Linux systems also, but I don't think Process Lasso is going there yet either.

For the most part, Intel doesn't need this. Just disable those useless "e-cores" and the L3 cache is shared with all the p-cores equally. Just have to make sure the number of tasks doesn't overflow that L3 cache.

Task pinning is generally only beneficial on AMD CPUs and multiple socket systems from both AMD and Intel.

With Intel multiple socket systems you can easily just run more than one BOINC instance, each one pinned to a specific socket. Then you don't have to change this ever again.

Additionally, you can do the same with AMD CPU setups. Run multiple instances each one pinned to a specific CCX then you never have to reconfigure anything again. On Zen3, I think and newer, you simply pin it to the CCD since each CCD only contains a single CCX. I only said CCX since they all have a CCX, but the older Zens have two CCX per one CCD and each CCX has it's own L3 cache domain. So to be safe, just pin each BOINC instance to a CCX and you're done.